Data for: Dynamic Reconstruction Based Representation Learning for Multivariable Process Monitoring

As a public benchmark of chemical industrial process, TEP is well suited for multivariable control problems. This simulation data can be downloaded from the website: http://web.mit.edu/braatzgroup/links.html The 41 measured variables, with the 22 continuous measurements are sampled with the sampling interval of 3mins, while the 19 composition variables are generated at time delays that vary from 6 to 15mins. 21 types of identified faults are inside. Each faulty state consists of 480 samples that are used as training dataset, and 960 samples are used as testing dataset with faults induced after 8 hours, which corresponds to 160 samples.作为化工过程的公开基准测试集，田纳西-伊斯曼过程（TEP）非常适用于多变量控制问题。该仿真数据集可从以下网址下载：http://web.mit.edu/braatzgroup/links.html。该数据集包含41个测量变量，其中22项为连续测量数据，采样间隔为3分钟；剩余19项成分变量的生成时延介于6至15分钟区间内。数据集内置21种已识别故障类型。每种故障状态均包含480个样本作为训练集，另有960个样本作为测试集；测试集的故障于运行8小时后注入，该时间点对应160个采样样本。

搜集汇总

背景与挑战

背景概述

该数据集基于田纳西伊士曼过程（TEP）化工过程基准，包含41个测量变量（22个连续测量和19个成分变量）以及21种故障类型，适用于多变量过程监控研究。数据采样时间间隔为3分钟，成分变量有6至15分钟的时间延迟，每个故障状态包含480个训练样本和960个测试样本，测试中故障在8小时后引入。